A novelty detection framework based on fuzzy entropy for a complex manufacturing process

Abstract

In this paper a new Novelty Detection framework is presented, which is created by taking advantage of the Fuzzy Entropy property of Fuzzy Logic systems. The framework’s aim is to create a linguistic-based feedback mechanism for advising the process users (Human-Centric System) on the performance of a complex manufacturing process. The manufacturing process under investigation is the Friction Stir Welding (FSW) process. The presented methodology comprises a data-driven model-based approach, which is the main facet of the framework. The proposed system has a Neural-Fuzzy structure that learns from process data to predict a number of process characteristics. Via the created Novelty Detection framework, we show how we can take advantage of the properties of the system to a) alert the user when a ‘novelty’ (i.e. new condition) appears in the system, and b) to advise the user in how reliable the system’s predictions are when the novelty occurs. The user feedback is provided in linguistic form by taking advantage the inherent features of the Fuzzy Logic– based approach. A number of simulations and experimental results based on a complex manufacturing case-study demonstrate the effectiveness and usefulness of the created Human-Centric System, which could be used as a form of decision support.